1 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
2 Alan Hayward <alan.hayward@arm.com>
3 David Sherwood <david.sherwood@arm.com>
5 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
6 so that both the length == 3 and length != 3 cases set up their
7 own permute vectors. Add comments explaining why we know the
8 number of elements is constant.
9 (vect_permute_load_chain): Likewise.
11 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
12 Alan Hayward <alan.hayward@arm.com>
13 David Sherwood <david.sherwood@arm.com>
15 * machmode.h (mode_nunits): Change from unsigned char to
17 (ONLY_FIXED_SIZE_MODES): New macro.
18 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
19 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
20 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
22 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
23 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
24 or if measurement_type is not polynomial.
25 * genmodes.c (ZERO_COEFFS): New macro.
26 (emit_mode_nunits_inline): Make mode_nunits_inline return a
28 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
29 Use ZERO_COEFFS when emitting initializers.
30 * data-streamer.h (bp_pack_poly_value): New function.
31 (bp_unpack_poly_value): Likewise.
32 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
34 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
36 * tree.c (make_vector_type): Remove temporary shim and make
37 the real function take the number of units as a poly_uint64
39 (build_vector_type_for_mode): Handle polynomial nunits.
40 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
41 * emit-rtl.c (const_vec_series_p_1): Likewise.
42 (gen_rtx_CONST_VECTOR): Likewise.
43 * fold-const.c (test_vec_duplicate_folding): Likewise.
44 * genrecog.c (validate_pattern): Likewise.
45 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
46 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
47 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
48 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
49 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
50 * rtlanal.c (subreg_get_info): Likewise.
51 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
52 (vect_grouped_load_supported): Likewise.
53 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
54 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
55 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
56 (simplify_const_unary_operation, simplify_binary_operation_1)
57 (simplify_const_binary_operation, simplify_ternary_operation)
58 (test_vector_ops_duplicate, test_vector_ops): Likewise.
59 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
60 instead of CONST_VECTOR_NUNITS.
61 * varasm.c (output_constant_pool_2): Likewise.
62 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
63 explicit-encoded elements in the XVEC for variable-length vectors.
65 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
67 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
69 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
70 Alan Hayward <alan.hayward@arm.com>
71 David Sherwood <david.sherwood@arm.com>
73 * coretypes.h (fixed_size_mode): Declare.
74 (fixed_size_mode_pod): New typedef.
75 * builtins.h (target_builtins::x_apply_args_mode)
76 (target_builtins::x_apply_result_mode): Change type to
78 * builtins.c (apply_args_size, apply_result_size, result_vector)
79 (expand_builtin_apply_args_1, expand_builtin_apply)
80 (expand_builtin_return): Update accordingly.
82 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
84 * cse.c (hash_rtx_cb): Hash only the encoded elements.
85 * cselib.c (cselib_hash_rtx): Likewise.
86 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
87 CONST_VECTOR encoding.
89 2017-01-03 Jakub Jelinek <jakub@redhat.com>
90 Jeff Law <law@redhat.com>
93 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
94 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
95 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
96 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
99 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
100 explicitly probe *sp in a noreturn function if there were any callee
101 register saves or frame pointer is needed.
103 2018-01-03 Jakub Jelinek <jakub@redhat.com>
106 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
107 BLKmode for ternary, binary or unary expressions.
110 * var-tracking.c (delete_vta_debug_insn): New inline function.
111 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
112 insns from get_insns () to NULL instead of each bb separately.
113 Use delete_vta_debug_insn. No longer static.
114 (vt_debug_insns_local, variable_tracking_main_1): Adjust
115 delete_vta_debug_insns callers.
116 * rtl.h (delete_vta_debug_insns): Declare.
117 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
118 instead of variable_tracking_main.
120 2018-01-03 Martin Sebor <msebor@redhat.com>
123 * doc/extend.texi (attribute const): Fix a typo.
124 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
125 issuing -Wsuggest-attribute for void functions.
127 2018-01-03 Martin Sebor <msebor@redhat.com>
129 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
130 offset_int::from instead of wide_int::to_shwi.
131 (maybe_diag_overlap): Remove assertion.
132 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
133 * gimple-ssa-sprintf.c (format_directive): Same.
134 (parse_directive): Same.
135 (sprintf_dom_walker::compute_format_length): Same.
136 (try_substitute_return_value): Same.
138 2017-01-03 Jeff Law <law@redhat.com>
141 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
142 non-constant residual for zero at runtime and avoid probing in
143 that case. Reorganize code for trailing problem to mirror handling
146 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
148 PR tree-optimization/83501
149 * tree-ssa-strlen.c (get_string_cst): New.
150 (handle_char_store): Call get_string_cst.
152 2018-01-03 Martin Liska <mliska@suse.cz>
154 PR tree-optimization/83593
155 * tree-ssa-strlen.c: Include tree-cfg.h.
156 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
157 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
158 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
160 (strlen_dom_walker::before_dom_children): Call
161 gimple_purge_dead_eh_edges. Dump tranformation with details
163 (strlen_dom_walker::before_dom_children): Update call by adding
164 new argument cleanup_eh.
165 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
167 2018-01-03 Martin Liska <mliska@suse.cz>
170 * cif-code.def (VARIADIC_THUNK): New enum value.
171 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
174 2018-01-03 Jan Beulich <jbeulich@suse.com>
176 * sse.md (mov<mode>_internal): Tighten condition for when to use
177 vmovdqu<ssescalarsize> for TI and OI modes.
179 2018-01-03 Jakub Jelinek <jakub@redhat.com>
181 Update copyright years.
183 2018-01-03 Martin Liska <mliska@suse.cz>
186 * ipa-visibility.c (function_and_variable_visibility): Skip
187 functions with noipa attribure.
189 2018-01-03 Jakub Jelinek <jakub@redhat.com>
191 * gcc.c (process_command): Update copyright notice dates.
192 * gcov-dump.c (print_version): Ditto.
193 * gcov.c (print_version): Ditto.
194 * gcov-tool.c (print_version): Ditto.
195 * gengtype.c (create_file): Ditto.
196 * doc/cpp.texi: Bump @copying's copyright year.
197 * doc/cppinternals.texi: Ditto.
198 * doc/gcc.texi: Ditto.
199 * doc/gccint.texi: Ditto.
200 * doc/gcov.texi: Ditto.
201 * doc/install.texi: Ditto.
202 * doc/invoke.texi: Ditto.
204 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
206 * vector-builder.h (vector_builder::m_full_nelts): Change from
207 unsigned int to poly_uint64.
208 (vector_builder::full_nelts): Update prototype accordingly.
209 (vector_builder::new_vector): Likewise.
210 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
211 (vector_builder::operator ==): Likewise.
212 (vector_builder::finalize): Likewise.
213 * int-vector-builder.h (int_vector_builder::int_vector_builder):
214 Take the number of elements as a poly_uint64 rather than an
216 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
217 from unsigned int to poly_uint64.
218 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
219 (vec_perm_indices::new_vector): Likewise.
220 (vec_perm_indices::length): Likewise.
221 (vec_perm_indices::nelts_per_input): Likewise.
222 (vec_perm_indices::input_nelts): Likewise.
223 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
224 number of elements per input as a poly_uint64 rather than an
225 unsigned int. Use the original encoding for variable-length
226 vectors, rather than clamping each individual element.
227 For the second and subsequent elements in each pattern,
228 clamp the step and base before clamping their sum.
229 (vec_perm_indices::series_p): Handle polynomial element counts.
230 (vec_perm_indices::all_in_range_p): Likewise.
231 (vec_perm_indices_to_tree): Likewise.
232 (vec_perm_indices_to_rtx): Likewise.
233 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
234 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
235 (tree_vector_builder::new_binary_operation): Handle polynomial
236 element counts. Return false if we need to know the number
237 of elements at compile time.
238 * fold-const.c (fold_vec_perm): Punt if the number of elements
239 isn't known at compile time.
241 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
243 * vec-perm-indices.h (vec_perm_builder): Change element type
244 from HOST_WIDE_INT to poly_int64.
245 (vec_perm_indices::element_type): Update accordingly.
246 (vec_perm_indices::clamp): Handle polynomial element_types.
247 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
248 (vec_perm_indices::all_in_range_p): Likewise.
249 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
251 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
252 polynomial vec_perm_indices element types.
253 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
254 * fold-const.c (fold_vec_perm): Likewise.
255 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
256 * tree-vect-generic.c (lower_vec_perm): Likewise.
257 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
258 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
259 element type to HOST_WIDE_INT.
261 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
262 Alan Hayward <alan.hayward@arm.com>
263 David Sherwood <david.sherwood@arm.com>
265 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
266 rather than an int. Use plus_constant.
267 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
268 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
270 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
271 Alan Hayward <alan.hayward@arm.com>
272 David Sherwood <david.sherwood@arm.com>
274 * calls.c (emit_call_1, expand_call): Change struct_value_size from
275 a HOST_WIDE_INT to a poly_int64.
277 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
278 Alan Hayward <alan.hayward@arm.com>
279 David Sherwood <david.sherwood@arm.com>
281 * calls.c (load_register_parameters): Cope with polynomial
282 mode sizes. Require a constant size for BLKmode parameters
283 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
284 forces a parameter to be padded at the lsb end in order to
285 fill a complete number of words, require the parameter size
286 to be ordered wrt UNITS_PER_WORD.
288 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
289 Alan Hayward <alan.hayward@arm.com>
290 David Sherwood <david.sherwood@arm.com>
292 * reload1.c (spill_stack_slot_width): Change element type
293 from unsigned int to poly_uint64_pod.
294 (alter_reg): Treat mode sizes as polynomial.
296 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
297 Alan Hayward <alan.hayward@arm.com>
298 David Sherwood <david.sherwood@arm.com>
300 * reload.c (complex_word_subreg_p): New function.
301 (reload_inner_reg_of_subreg, push_reload): Use it.
303 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
304 Alan Hayward <alan.hayward@arm.com>
305 David Sherwood <david.sherwood@arm.com>
307 * lra-constraints.c (process_alt_operands): Reject matched
308 operands whose sizes aren't ordered.
309 (match_reload): Refer to this check here.
311 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
312 Alan Hayward <alan.hayward@arm.com>
313 David Sherwood <david.sherwood@arm.com>
315 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
316 that the mode size is in the set {1, 2, 4, 8, 16}.
318 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
319 Alan Hayward <alan.hayward@arm.com>
320 David Sherwood <david.sherwood@arm.com>
322 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
323 Use plus_constant instead of gen_rtx_PLUS.
325 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
326 Alan Hayward <alan.hayward@arm.com>
327 David Sherwood <david.sherwood@arm.com>
329 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
330 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
331 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
332 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
333 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
334 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
335 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
336 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
337 * config/i386/i386.c (ix86_push_rounding): ...this new function.
338 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
340 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
341 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
342 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
343 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
344 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
345 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
346 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
347 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
348 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
349 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
351 * expr.c (emit_move_resolve_push): Treat the input and result
352 of PUSH_ROUNDING as a poly_int64.
353 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
354 (emit_push_insn): Likewise.
355 * lra-eliminations.c (mark_not_eliminable): Likewise.
356 * recog.c (push_operand): Likewise.
357 * reload1.c (elimination_effects): Likewise.
358 * rtlanal.c (nonzero_bits1): Likewise.
359 * calls.c (store_one_arg): Likewise. Require the padding to be
360 known at compile time.
362 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
363 Alan Hayward <alan.hayward@arm.com>
364 David Sherwood <david.sherwood@arm.com>
366 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
367 Use plus_constant instead of gen_rtx_PLUS.
369 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
370 Alan Hayward <alan.hayward@arm.com>
371 David Sherwood <david.sherwood@arm.com>
373 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
376 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
377 Alan Hayward <alan.hayward@arm.com>
378 David Sherwood <david.sherwood@arm.com>
380 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
381 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
382 via stack temporaries. Treat the mode size as polynomial too.
384 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
385 Alan Hayward <alan.hayward@arm.com>
386 David Sherwood <david.sherwood@arm.com>
388 * expr.c (expand_expr_real_2): When handling conversions involving
389 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
390 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
391 as a poly_uint64 too.
393 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
394 Alan Hayward <alan.hayward@arm.com>
395 David Sherwood <david.sherwood@arm.com>
397 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
399 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
400 Alan Hayward <alan.hayward@arm.com>
401 David Sherwood <david.sherwood@arm.com>
403 * combine.c (can_change_dest_mode): Handle polynomial
404 REGMODE_NATURAL_SIZE.
405 * expmed.c (store_bit_field_1): Likewise.
406 * expr.c (store_constructor): Likewise.
407 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
408 and polynomial REGMODE_NATURAL_SIZE.
409 (gen_lowpart_common): Likewise.
410 * reginfo.c (record_subregs_of_mode): Likewise.
411 * rtlanal.c (read_modify_subreg_p): Likewise.
413 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
414 Alan Hayward <alan.hayward@arm.com>
415 David Sherwood <david.sherwood@arm.com>
417 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
420 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
421 Alan Hayward <alan.hayward@arm.com>
422 David Sherwood <david.sherwood@arm.com>
424 * match.pd: Cope with polynomial numbers of vector elements.
426 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
427 Alan Hayward <alan.hayward@arm.com>
428 David Sherwood <david.sherwood@arm.com>
430 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
431 in a POINTER_PLUS_EXPR.
433 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
434 Alan Hayward <alan.hayward@arm.com>
435 David Sherwood <david.sherwood@arm.com>
437 * omp-simd-clone.c (simd_clone_subparts): New function.
438 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
439 (ipa_simd_modify_function_body): Likewise.
441 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
442 Alan Hayward <alan.hayward@arm.com>
443 David Sherwood <david.sherwood@arm.com>
445 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
446 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
447 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
448 (expand_vector_condition, vector_element): Likewise.
449 (subparts_gt): New function.
450 (get_compute_type): Use subparts_gt.
451 (count_type_subparts): Delete.
452 (expand_vector_operations_1): Use subparts_gt instead of
455 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
456 Alan Hayward <alan.hayward@arm.com>
457 David Sherwood <david.sherwood@arm.com>
459 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
460 (vect_compile_time_alias): ...this new function. Do the calculation
461 on poly_ints rather than trees.
462 (vect_prune_runtime_alias_test_list): Update call accordingly.
464 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
465 Alan Hayward <alan.hayward@arm.com>
466 David Sherwood <david.sherwood@arm.com>
468 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
470 (vect_schedule_slp_instance): Likewise.
472 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
473 Alan Hayward <alan.hayward@arm.com>
474 David Sherwood <david.sherwood@arm.com>
476 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
477 constant and extern definitions for variable-length vectors.
478 (vect_get_constant_vectors): Note that the number of units
479 is known to be constant.
481 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
482 Alan Hayward <alan.hayward@arm.com>
483 David Sherwood <david.sherwood@arm.com>
485 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
486 of units as polynomial. Choose between WIDE and NARROW based
489 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
490 Alan Hayward <alan.hayward@arm.com>
491 David Sherwood <david.sherwood@arm.com>
493 * tree-vect-stmts.c (simd_clone_subparts): New function.
494 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
496 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
497 Alan Hayward <alan.hayward@arm.com>
498 David Sherwood <david.sherwood@arm.com>
500 * tree-vect-stmts.c (vectorizable_call): Treat the number of
501 vectors as polynomial. Use build_index_vector for
504 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
505 Alan Hayward <alan.hayward@arm.com>
506 David Sherwood <david.sherwood@arm.com>
508 * tree-vect-stmts.c (get_load_store_type): Treat the number of
509 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
510 for variable-length vectors.
511 (vectorizable_mask_load_store): Treat the number of units as
512 polynomial, asserting that it is constant if the condition has
513 already been enforced.
514 (vectorizable_store, vectorizable_load): Likewise.
516 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
517 Alan Hayward <alan.hayward@arm.com>
518 David Sherwood <david.sherwood@arm.com>
520 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
521 of units as polynomial. Punt if we can't tell at compile time
522 which vector contains the final result.
524 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
525 Alan Hayward <alan.hayward@arm.com>
526 David Sherwood <david.sherwood@arm.com>
528 * tree-vect-loop.c (vectorizable_induction): Treat the number
529 of units as polynomial. Punt on SLP inductions. Use an integer
530 VEC_SERIES_EXPR for variable-length integer reductions. Use a
531 cast of such a series for variable-length floating-point
534 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
535 Alan Hayward <alan.hayward@arm.com>
536 David Sherwood <david.sherwood@arm.com>
538 * tree.h (build_index_vector): Declare.
539 * tree.c (build_index_vector): New function.
540 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
541 of units as polynomial, forcibly converting it to a constant if
542 vectorizable_reduction has already enforced the condition.
543 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
544 to create a {1,2,3,...} vector.
545 (vectorizable_reduction): Treat the number of units as polynomial.
546 Choose vectype_in based on the largest scalar element size rather
547 than the smallest number of units. Enforce the restrictions
550 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
551 Alan Hayward <alan.hayward@arm.com>
552 David Sherwood <david.sherwood@arm.com>
554 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
555 number of units as polynomial.
557 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
558 Alan Hayward <alan.hayward@arm.com>
559 David Sherwood <david.sherwood@arm.com>
561 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
562 * target.def (autovectorize_vector_sizes): Return the vector sizes
563 by pointer, using vector_sizes rather than a bitmask.
564 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
565 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
566 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
568 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
569 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
570 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
571 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
572 * omp-general.c (omp_max_vf): Likewise.
573 * omp-low.c (omp_clause_aligned_alignment): Likewise.
574 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
575 * tree-vect-loop.c (vect_analyze_loop): Likewise.
576 * tree-vect-slp.c (vect_slp_bb): Likewise.
577 * doc/tm.texi: Regenerate.
578 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
580 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
581 the vector size as a poly_uint64 rather than an unsigned int.
582 (current_vector_size): Change from an unsigned int to a poly_uint64.
583 (get_vectype_for_scalar_type): Update accordingly.
584 * tree.h (build_truth_vector_type): Take the size and number of
585 units as a poly_uint64 rather than an unsigned int.
586 (build_vector_type): Add a temporary overload that takes
587 the number of units as a poly_uint64 rather than an unsigned int.
588 * tree.c (make_vector_type): Likewise.
589 (build_truth_vector_type): Take the number of units as a poly_uint64
590 rather than an unsigned int.
592 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
593 Alan Hayward <alan.hayward@arm.com>
594 David Sherwood <david.sherwood@arm.com>
596 * target.def (get_mask_mode): Take the number of units and length
597 as poly_uint64s rather than unsigned ints.
598 * targhooks.h (default_get_mask_mode): Update accordingly.
599 * targhooks.c (default_get_mask_mode): Likewise.
600 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
601 * doc/tm.texi: Regenerate.
603 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
604 Alan Hayward <alan.hayward@arm.com>
605 David Sherwood <david.sherwood@arm.com>
607 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
608 * omp-general.c (omp_max_vf): Likewise.
609 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
610 (expand_omp_simd): Handle polynomial safelen.
611 * omp-low.c (omplow_simd_context): Add a default constructor.
612 (omplow_simd_context::max_vf): Change from int to poly_uint64.
613 (lower_rec_simd_input_clauses): Update accordingly.
614 (lower_rec_input_clauses): Likewise.
616 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
617 Alan Hayward <alan.hayward@arm.com>
618 David Sherwood <david.sherwood@arm.com>
620 * tree-vectorizer.h (vect_nunits_for_cost): New function.
621 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
622 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
623 (vect_analyze_slp_cost): Likewise.
624 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
625 (vect_model_load_cost): Likewise.
627 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
628 Alan Hayward <alan.hayward@arm.com>
629 David Sherwood <david.sherwood@arm.com>
631 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
632 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
633 from an unsigned int * to a poly_uint64_pod *.
634 (calculate_unrolling_factor): New function.
635 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
637 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
638 Alan Hayward <alan.hayward@arm.com>
639 David Sherwood <david.sherwood@arm.com>
641 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
642 from an unsigned int to a poly_uint64.
643 (_loop_vec_info::slp_unrolling_factor): Likewise.
644 (_loop_vec_info::vectorization_factor): Change from an int
646 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
647 (vect_get_num_vectors): New function.
648 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
649 (vect_get_num_copies): Use vect_get_num_vectors.
650 (vect_analyze_data_ref_dependences): Change max_vf from an int *
651 to an unsigned int *.
652 (vect_analyze_data_refs): Change min_vf from an int * to a
654 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
655 than an unsigned HOST_WIDE_INT.
656 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
657 (vect_analyze_data_ref_dependence): Change max_vf from an int *
658 to an unsigned int *.
659 (vect_analyze_data_ref_dependences): Likewise.
660 (vect_compute_data_ref_alignment): Handle polynomial vf.
661 (vect_enhance_data_refs_alignment): Likewise.
662 (vect_prune_runtime_alias_test_list): Likewise.
663 (vect_shift_permute_load_chain): Likewise.
664 (vect_supportable_dr_alignment): Likewise.
665 (dependence_distance_ge_vf): Take the vectorization factor as a
666 poly_uint64 rather than an unsigned HOST_WIDE_INT.
667 (vect_analyze_data_refs): Change min_vf from an int * to a
669 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
670 vfm1 as a poly_uint64 rather than an int. Make the same change
671 for the returned bound_scalar.
672 (vect_gen_vector_loop_niters): Handle polynomial vf.
673 (vect_do_peeling): Likewise. Update call to
674 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
675 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
677 * tree-vect-loop.c (vect_determine_vectorization_factor)
678 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
679 (vect_get_known_peeling_cost): Likewise.
680 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
681 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
682 (vect_transform_loop): Likewise. Use the lowest possible VF when
683 updating the upper bounds of the loop.
684 (vect_min_worthwhile_factor): Make static. Return an unsigned int
686 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
687 polynomial unroll factors.
688 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
689 (vect_make_slp_decision): Likewise.
690 (vect_supported_load_permutation_p): Likewise, and polynomial
692 (vect_analyze_slp_cost): Handle polynomial vf.
693 (vect_slp_analyze_node_operations): Likewise.
694 (vect_slp_analyze_bb_1): Likewise.
695 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
696 than an unsigned HOST_WIDE_INT.
697 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
698 (vectorizable_load): Handle polynomial vf.
699 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
701 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
703 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
704 Alan Hayward <alan.hayward@arm.com>
705 David Sherwood <david.sherwood@arm.com>
707 * match.pd: Handle bit operations involving three constants
708 and try to fold one pair.
710 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
712 * tree-vect-loop-manip.c: Include gimple-fold.h.
713 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
714 niters_maybe_zero parameters. Handle other cases besides a step of 1.
715 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
716 Add a path that uses a step of VF instead of 1, but disable it
718 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
719 and niters_no_overflow parameters. Update calls to
720 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
721 Create a new SSA name if the latter choses to use a ste other
722 than zero, and return it via niters_vector_mult_vf_var.
723 * tree-vect-loop.c (vect_transform_loop): Update calls to
724 vect_do_peeling, vect_gen_vector_loop_niters and
725 slpeel_make_loop_iterate_ntimes.
726 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
727 (vect_gen_vector_loop_niters): Update declarations after above changes.
729 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
731 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
732 128-bit round to integer instructions.
733 (ceil<mode>2): Likewise.
734 (btrunc<mode>2): Likewise.
735 (round<mode>2): Likewise.
737 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
739 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
740 unaligned VSX load/store on P8/P9.
741 (expand_block_clear): Allow the use of unaligned VSX
744 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
746 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
748 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
749 swap associated with both a load and a store.
751 2018-01-02 Andrew Waterman <andrew@sifive.com>
753 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
754 * config/riscv/riscv.md (clear_cache): Use it.
756 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
758 * web.c: Remove out-of-date comment.
760 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
762 * expr.c (fixup_args_size_notes): Check that any existing
763 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
764 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
765 (emit_single_push_insn): ...here.
767 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
769 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
770 (const_vector_encoded_nelts): New function.
771 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
772 (const_vector_int_elt, const_vector_elt): Declare.
773 * emit-rtl.c (const_vector_int_elt_1): New function.
774 (const_vector_elt): Likewise.
775 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
778 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
780 * expr.c: Include rtx-vector-builder.h.
781 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
782 directly on the tree encoding.
783 (const_vector_from_tree): Likewise.
784 * optabs.c: Include rtx-vector-builder.h.
785 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
786 sequence of "u" values.
787 * vec-perm-indices.c: Include rtx-vector-builder.h.
788 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
789 directly on the vec_perm_indices encoding.
791 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
793 * doc/rtl.texi (const_vector): Describe new encoding scheme.
794 * Makefile.in (OBJS): Add rtx-vector-builder.o.
795 * rtx-vector-builder.h: New file.
796 * rtx-vector-builder.c: Likewise.
797 * rtl.h (rtx_def::u2): Add a const_vector field.
798 (CONST_VECTOR_NPATTERNS): New macro.
799 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
800 (CONST_VECTOR_DUPLICATE_P): Likewise.
801 (CONST_VECTOR_STEPPED_P): Likewise.
802 (CONST_VECTOR_ENCODED_ELT): Likewise.
803 (const_vec_duplicate_p): Check for a duplicated vector encoding.
804 (unwrap_const_vec_duplicate): Likewise.
805 (const_vec_series_p): Check for a non-duplicated vector encoding.
806 Say that the function only returns true for integer vectors.
807 * emit-rtl.c: Include rtx-vector-builder.h.
808 (gen_const_vec_duplicate_1): Delete.
809 (gen_const_vector): Call gen_const_vec_duplicate instead of
810 gen_const_vec_duplicate_1.
811 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
812 (gen_const_vec_duplicate): Use rtx_vector_builder.
813 (gen_const_vec_series): Likewise.
814 (gen_rtx_CONST_VECTOR): Likewise.
815 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
816 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
817 Build a new vector rather than modifying a CONST_VECTOR in-place.
818 (handle_special_swappables): Update call accordingly.
819 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
820 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
821 Build a new vector rather than modifying a CONST_VECTOR in-place.
822 (handle_special_swappables): Update call accordingly.
824 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
826 * simplify-rtx.c (simplify_const_binary_operation): Use
827 CONST_VECTOR_ELT instead of XVECEXP.
829 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
831 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
832 the selector elements to be different from the data elements
833 if the selector is a VECTOR_CST.
834 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
835 ssizetype for the selector.
837 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
839 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
840 before testing each element individually.
841 * tree-vect-generic.c (lower_vec_perm): Likewise.
843 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
845 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
846 * selftest-run-tests.c (selftest::run_tests): Call it.
847 * vector-builder.h (vector_builder::operator ==): New function.
848 (vector_builder::operator !=): Likewise.
849 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
850 (vec_perm_indices::all_from_input_p): New function.
851 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
852 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
853 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
854 instead of reading the VECTOR_CST directly. Detect whether both
855 vector inputs are the same before constructing the vec_perm_indices,
856 and update the number of inputs argument accordingly. Use the
857 utility functions added above. Only construct sel2 if we need to.
859 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
861 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
862 the broadcast of the low byte.
863 (expand_mult_highpart): Use an explicit encoding for the permutes.
864 * optabs-query.c (can_mult_highpart_p): Likewise.
865 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
866 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
867 (vectorizable_bswap): Likewise.
868 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
869 explicit encoding for the power-of-2 permutes.
870 (vect_permute_store_chain): Likewise.
871 (vect_grouped_load_supported): Likewise.
872 (vect_permute_load_chain): Likewise.
874 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
876 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
877 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
878 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
879 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
880 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
881 (vect_gen_perm_mask_any): Likewise.
883 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
885 * int-vector-builder.h: New file.
886 * vec-perm-indices.h: Include int-vector-builder.h.
887 (vec_perm_indices): Redefine as an int_vector_builder.
888 (auto_vec_perm_indices): Delete.
889 (vec_perm_builder): Redefine as a stand-alone class.
890 (vec_perm_indices::vec_perm_indices): New function.
891 (vec_perm_indices::clamp): Likewise.
892 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
893 (vec_perm_indices::new_vector): New function.
894 (vec_perm_indices::new_expanded_vector): Update for new
895 vec_perm_indices class.
896 (vec_perm_indices::rotate_inputs): New function.
897 (vec_perm_indices::all_in_range_p): Operate directly on the
898 encoded form, without computing elided elements.
899 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
900 encoding. Update for new vec_perm_indices class.
901 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
902 the given vec_perm_builder.
903 (expand_vec_perm_var): Update vec_perm_builder constructor.
904 (expand_mult_highpart): Use vec_perm_builder instead of
905 auto_vec_perm_indices.
906 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
907 vec_perm_indices instead of auto_vec_perm_indices. Use a single
908 or double series encoding as appropriate.
909 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
910 vec_perm_indices instead of auto_vec_perm_indices.
911 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
912 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
913 (vect_permute_store_chain): Likewise.
914 (vect_grouped_load_supported): Likewise.
915 (vect_permute_load_chain): Likewise.
916 (vect_shift_permute_load_chain): Likewise.
917 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
918 (vect_transform_slp_perm_load): Likewise.
919 (vect_schedule_slp_instance): Likewise.
920 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
921 (vectorizable_mask_load_store): Likewise.
922 (vectorizable_bswap): Likewise.
923 (vectorizable_store): Likewise.
924 (vectorizable_load): Likewise.
925 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
926 vec_perm_indices instead of auto_vec_perm_indices. Use
927 tree_to_vec_perm_builder to read the vector from a tree.
928 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
929 vec_perm_builder instead of a vec_perm_indices.
930 (have_whole_vector_shift): Use vec_perm_builder and
931 vec_perm_indices instead of auto_vec_perm_indices. Leave the
932 truncation to calc_vec_perm_mask_for_shift.
933 (vect_create_epilog_for_reduction): Likewise.
934 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
935 from auto_vec_perm_indices to vec_perm_indices.
936 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
937 instead of changing individual elements.
938 (aarch64_vectorize_vec_perm_const): Use new_vector to install
939 the vector in d.perm.
940 * config/arm/arm.c (expand_vec_perm_d::perm): Change
941 from auto_vec_perm_indices to vec_perm_indices.
942 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
943 instead of changing individual elements.
944 (arm_vectorize_vec_perm_const): Use new_vector to install
945 the vector in d.perm.
946 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
947 Update vec_perm_builder constructor.
948 (rs6000_expand_interleave): Likewise.
949 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
950 (rs6000_expand_interleave): Likewise.
952 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
954 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
955 to qimode could truncate the indices.
956 * optabs.c (expand_vec_perm_var): Likewise.
958 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
960 * Makefile.in (OBJS): Add vec-perm-indices.o.
961 * vec-perm-indices.h: New file.
962 * vec-perm-indices.c: Likewise.
963 * target.h (vec_perm_indices): Replace with a forward class
965 (auto_vec_perm_indices): Move to vec-perm-indices.h.
966 * optabs.h: Include vec-perm-indices.h.
967 (expand_vec_perm): Delete.
968 (selector_fits_mode_p, expand_vec_perm_var): Declare.
969 (expand_vec_perm_const): Declare.
970 * target.def (vec_perm_const_ok): Replace with...
971 (vec_perm_const): ...this new hook.
972 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
973 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
974 * doc/tm.texi: Regenerate.
975 * optabs.def (vec_perm_const): Delete.
976 * doc/md.texi (vec_perm_const): Likewise.
977 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
978 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
979 expand_vec_perm for constant permutation vectors. Assert that
980 the mode of variable permutation vectors is the integer equivalent
981 of the mode that is being permuted.
982 * optabs-query.h (selector_fits_mode_p): Declare.
983 * optabs-query.c: Include vec-perm-indices.h.
984 (selector_fits_mode_p): New function.
985 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
986 is defined, instead of checking whether the vec_perm_const_optab
987 exists. Use targetm.vectorize.vec_perm_const instead of
988 targetm.vectorize.vec_perm_const_ok. Check whether the indices
989 fit in the vector mode before using a variable permute.
990 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
991 vec_perm_indices instead of an rtx.
992 (expand_vec_perm): Replace with...
993 (expand_vec_perm_const): ...this new function. Take the selector
994 as a vec_perm_indices rather than an rtx. Also take the mode of
995 the selector. Update call to shift_amt_for_vec_perm_mask.
996 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
997 Use vec_perm_indices::new_expanded_vector to expand the original
998 selector into bytes. Check whether the indices fit in the vector
999 mode before using a variable permute.
1000 (expand_vec_perm_var): Make global.
1001 (expand_mult_highpart): Use expand_vec_perm_const.
1002 * fold-const.c: Includes vec-perm-indices.h.
1003 * tree-ssa-forwprop.c: Likewise.
1004 * tree-vect-data-refs.c: Likewise.
1005 * tree-vect-generic.c: Likewise.
1006 * tree-vect-loop.c: Likewise.
1007 * tree-vect-slp.c: Likewise.
1008 * tree-vect-stmts.c: Likewise.
1009 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
1011 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
1012 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
1013 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
1014 (aarch64_vectorize_vec_perm_const): ...this new function.
1015 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
1016 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1017 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
1018 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
1019 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
1020 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1021 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
1023 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
1024 check for NEON modes.
1025 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
1026 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
1027 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
1028 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
1030 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
1031 the old VEC_PERM_CONST conditions.
1032 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
1033 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
1034 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
1035 (ia64_vectorize_vec_perm_const_ok): Merge into...
1036 (ia64_vectorize_vec_perm_const): ...this new function.
1037 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
1038 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
1039 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
1040 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
1041 * config/mips/mips.c (mips_expand_vec_perm_const)
1042 (mips_vectorize_vec_perm_const_ok): Merge into...
1043 (mips_vectorize_vec_perm_const): ...this new function.
1044 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
1045 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
1046 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
1047 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
1048 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
1049 (rs6000_expand_vec_perm_const): Delete.
1050 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
1052 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1053 (altivec_expand_vec_perm_const_le): Take each operand individually.
1054 Operate on constant selectors rather than rtxes.
1055 (altivec_expand_vec_perm_const): Likewise. Update call to
1056 altivec_expand_vec_perm_const_le.
1057 (rs6000_expand_vec_perm_const): Delete.
1058 (rs6000_vectorize_vec_perm_const_ok): Delete.
1059 (rs6000_vectorize_vec_perm_const): New function.
1060 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
1061 an element count and rtx array.
1062 (rs6000_expand_extract_even): Update call accordingly.
1063 (rs6000_expand_interleave): Likewise.
1064 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
1065 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
1066 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
1067 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
1068 (rs6000_expand_vec_perm_const): Delete.
1069 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
1070 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1071 (altivec_expand_vec_perm_const_le): Take each operand individually.
1072 Operate on constant selectors rather than rtxes.
1073 (altivec_expand_vec_perm_const): Likewise. Update call to
1074 altivec_expand_vec_perm_const_le.
1075 (rs6000_expand_vec_perm_const): Delete.
1076 (rs6000_vectorize_vec_perm_const_ok): Delete.
1077 (rs6000_vectorize_vec_perm_const): New function. Remove stray
1078 reference to the SPE evmerge intructions.
1079 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
1080 an element count and rtx array.
1081 (rs6000_expand_extract_even): Update call accordingly.
1082 (rs6000_expand_interleave): Likewise.
1083 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
1084 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
1086 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1088 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1090 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
1091 vector mode and that that mode matches the mode of the data
1093 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
1094 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
1095 directly using expand_vec_perm_1 when forcing selectors into
1097 (expand_vec_perm_var): New function, split out from expand_vec_perm.
1099 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1101 * optabs-query.h (can_vec_perm_p): Delete.
1102 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
1103 * optabs-query.c (can_vec_perm_p): Split into...
1104 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
1105 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
1106 particular selector is valid.
1107 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
1108 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
1109 (vect_grouped_load_supported): Likewise.
1110 (vect_shift_permute_load_chain): Likewise.
1111 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
1112 (vect_transform_slp_perm_load): Likewise.
1113 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
1114 (vectorizable_bswap): Likewise.
1115 (vect_gen_perm_mask_checked): Likewise.
1116 * fold-const.c (fold_ternary_loc): Likewise. Don't take
1117 implementations of variable permutation vectors into account
1118 when deciding which selector to use.
1119 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
1120 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
1121 with a false third argument.
1122 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
1123 to test whether the constant selector is valid and can_vec_perm_var_p
1124 to test whether a variable selector is valid.
1126 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1128 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
1129 * optabs-query.c (can_vec_perm_p): Likewise.
1130 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
1131 instead of vec_perm_indices.
1132 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
1133 (vect_gen_perm_mask_checked): Likewise,
1134 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
1135 (vect_gen_perm_mask_checked): Likewise,
1137 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
1139 * optabs-query.h (qimode_for_vec_perm): Declare.
1140 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
1141 (qimode_for_vec_perm): ...this new function.
1142 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
1144 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
1146 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
1147 does not have a conditional at the top.
1149 2018-01-02 Richard Biener <rguenther@suse.de>
1151 * ipa-inline.c (big_speedup_p): Fix expression.
1153 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
1156 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
1159 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
1163 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
1164 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
1165 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
1166 cond_taken_branch_cost 3->4.
1168 2018-01-01 Jakub Jelinek <jakub@redhat.com>
1170 PR tree-optimization/83581
1171 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
1172 TODO_cleanup_cfg if any changes have been made.
1175 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
1176 convert_modes if target mode has the right side, but different mode
1180 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
1181 last argument when extracting from CONCAT. If either from_real or
1182 from_imag is NULL, use expansion through memory. If result is not
1183 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
1184 the parts directly to inner mode, if even that fails, use expansion
1188 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
1189 check for bswap in mode rather than HImode and use that in expand_unop
1192 Copyright (C) 2018 Free Software Foundation, Inc.
1194 Copying and distribution of this file, with or without modification,
1195 are permitted in any medium without royalty provided the copyright
1196 notice and this notice are preserved.